Selective auditory attention decoding in bilateral cochlear implant users to music instruments

Objective.electroencephalography (EEG) data can be used to decode an attended sound source in normal-hearing (NH) listeners, even for music stimuli. This information could steer the sound processing strategy for cochlear implants (CIs) users, potentially improving their music listening experience. The aim of this study was to investigate whether selective auditory attention decoding (SAAD) could be performed in CI users for music stimuli.Approach.High-density EEG was recorded from 8 NH listeners and 8 CI users. Duets containing a clarinet and cello were dichotically presented. A linear decoder was trained to reconstruct audio features of the attended instrument from EEG data. The estimated attended instrument was selected based on which of the two instruments had a higher correlation to the reconstructed instrument. EEG recordings are challenging in CI users, as these devices introduce strong electrical artifacts. In this work we also propose a new artifact rejection technique that employs independent component analysis (ICA) to calculate independent components and to automate their selection for removal, which we termed Artifact suppression ICA.Main results.In this paper, we showed that it was possible to perform SAAD for music in CI users. The decoding accuracies were 59.4% for NH listeners and 59.8% for CI users with the proposed algorithm. Using the proposed algorithm, the correlation coefficients between the reconstructed audio feature and the attended audio feature were improved in conditions where the artifact was more prominentSignificance.The results indicate that selective auditory attention to musical instruments can be effectively decoded, and that this decoding is enhanced by the new artifact reduction algorithm, particularly in scenarios where the cochlear implant's electrical artifact has greater influence. Moreover, these results could be relevant as an objective measure of music perception or for a brain computer interface that improves music enjoyment. Additionally we showed that the stimulation artifact can be suppressed. The ethics of MHH approved this study (8874_BO_K_2020).